Machine Learning Approach to Detect Red-Eye Using Pixel Detection Technique

摘要

Sometimes in Flash Photography red colored patches occurred in human eyes. It is actually a reflection of bright flash light reflected from blood vessels in the eyes, giving the eye an unnatural red hue. Red-eye is a big problem in professional photography. Most red-eye reduction systems in many editing software needed the user to identify the red-eye and make an outline through the red-eye. Here we propose an Automatic Red-Eye Detection System instead. The system contains a red-eye detector that finds bunch of red pixels those are clustered to gather, a state of face detector that used to eliminate most false positives (pixel clusters that look red eyes but are not); and a redeye outline detector. All three detectors are automatically learned from the taken datasets and with a proper classifiers using boosting. For creating a fully Automatic Red-Eye Corrector this system needed to be combined with a functional Red-Eye Reduction model.